DISCUSSION
Our overall results demonstrate rapid evolution in three out of ten traits under in situ climate manipulations in natural plant communities after merely 10 years, i.e. at most 10 generations of our annual study species. This is a remarkably short time span, given that numerous interacting factors may hamper evolution in natural communities (Hoffmann & Sgró 2011; Shaw & Etterson 2012). The fact that this evolutionary response was consistent in two independent sites renders chance effects, e.g. genetic drift, unlikely to cause these results and underpins that the evolutionary response was directly driven by manipulated rainfall. Intriguingly, our multiple independent lines of evidence corroborate that these changes were adaptive.
After 10 years of artificial drought, phenology had evolved both in chronological (days to flowering) and ontogenetic (leaf number at flowering) time. Theory suggests accelerated life-cycles as a drought avoidance strategy that reduces the risk of mortality before reproduction (Cohen 1976; Kigel et al. 2011), albeit at the cost of smaller plant size and hence possibly lower competitive ability (Liancourt & Tielbörger 2009; Kigel et al. 2011). In line with theory, plants from dry-manipulated plots flowered earlier and with fewer leaves than plants from control and wet plots. Moreover, this rapid evolutionary response paralleled the long-term evolutionary response of B. didyma along the natural rainfall gradient where plants from more arid sites flowered earlier; a trend found in many other annuals along natural rainfall gradients (Kigel et al . 2011; Wolfe & Tonsor 2014; Kurze et al . 2017). Interestingly, the observed 3-4 days acceleration in phenology corresponds to an ecological distance of c. 100 mm lower rainfall at origin for annuals along our study gradient (Kigel et al . 2011; Kurzeet al. 2017). Given the magnitude of rainfall reduction in dry plots (-90 mm in SA, -160 mm in M), this suggests that phenology could actually track a substantial part of the imposed change in rainfall. The adaptivity of accelerated phenology under drought was furthermore corroborated by our selection analyses under controlled, unconfounded (Mitchell-Olds & Schmitt 2006) watering conditions in the greenhouse. Here, earlier flowering with fewer leaves was stronger favored under low than under high water availability. These multiple lines of evidence – theory, natural rainfall gradient, selection analyses, and consistency in both sites – provide compelling evidence that the observed rapid evolution in phenology was adaptive.
Rapid evolution of earlier phenology under drought was previously reported from a Californian climate manipulation site (Nguyen et al. 2016) and from resurrection studies (Franks et al. 2007; Vigouroux et al. 2011; Nevo et al. 2012; Hamann et al. 2018). If phenology was reported, no evolution occurred merely in one perennial (Ravenscroft et al. 2014) or under elevated CO2 (Grossman & Rice 2014). Therefore, phenology appears a key trait for rapid drought adaptation in annuals, congruent with similar suggestions by theory and gradient studies (Cohen 1976; Kigel et al . 2011; Kurze et al. 2017). These findings may also indicate that phenology evolves easier than other, possibly more complex traits. However, more multi-trait studies (e.g. Ravenscroftet al. 2014; Nguyen et al. 2016; Hamann et al.2018) assessing comparable trait-sets are required to confirm this idea.
Here, we also observed rapid evolution in reproductive allocation. As competition is reduced in drier sites along our gradient (Schiffers & Tielbörger 2006), theory suggests reduced investment in vegetative tissue for outgrowing neighbors and increased allocation to reproduction (Aronson et al. 1990; 1993). In line with theory and in both sites, plants from dry manipulated plots produced 10-15% more seeds per vegetative biomass than control plants. Although reproductive allocation was rarely assessed in climate manipulation studies, a similar tendency was reported for a perennial herb (Ravenscroft et al . 2014). This evolutionary response was again congruent with our selection analyses in the greenhouse, and with the clinal trend in reproductive allocation along our natural rainfall gradient, and parallel clines in other species (summarized in Kurze et al. 2017). Thus, in all traits showing rapid evolution in the field, our independent lines of evidence demonstrate that these changes were adaptive. Intriguingly, parallel studies found that many plant community parameters were remarkably resistant to our climate manipulations (Tielbörger et al . 2014; Bilton et al . 2016). Though we have studied only a single species, our current findings suggest that rapid adaptive evolution possibly contributed to increasing population-level and community-level resistance to climate change.
Interestingly, evolutionary changes occurred solely in the dry manipulated plots, i.e. the treatment which increased, rather than decreased stress for resident plants. Drought likely lead to direct, rapid exclusion of drought-sensitive and late-flowering genotypes, especially in dry years. In wet plots, evolution may be slower because selection was likely driven by competition for additional resources (Schiffers & Tielbörger 2006) which causes smaller fitness differences, as was shown by cross-transplants with B. didyma (Ariza & Tielbörger 2011).
Despite the evidence for rapid adaptive evolution, seven further traits did not evolve. This was surprising because five of them exhibited clinal shifts along the rainfall gradient, suggesting that they contribute to B. didyma’ s long-term evolutionary response to drier climates: germination fraction, stomata density, height, vegetative biomass and seed number. In conjunction with existing theory we had expected corresponding evolution of these traits under climate manipulations (Westoby 1998; Liu et al. 2012; Tielbörger et al. 2012; ten Brink et al. 2020). Selection analyses supported this expectation for vegetative biomass, although not for stomata density and height; no tests were possible for germination fraction (no differential watering) and seed number (response variable in selection analyses). While empirical studies usually focused on (few) traits exhibiting rapid evolution, non-evolving traits have been reported before (e.g. Franks 2011; Ravenscroft et al. 2014; Nguyenet al. 2016). One possible explanation for the lack of evolution in some candidate traits is that selection on them was weakened by adaptation of the fast-evolving traits, i.e. evolution of further traits was unnecessary. Alternatively, the multiple potential constraints for evolution under natural conditions hindered adaptation in other traits, e.g. low genetic variation, genetic covariance and trade-offs among traits (Hoffmann & Sgró 2011; Shaw & Etterson 2012). In our case, negative genetic covariance potentially hindered evolution in vegetative biomass (Appendix Fig. S3). The observed rapid evolution in only a subset of traits may therefore indicate incomplete adaptation to new conditions, cautioning that climate change may imperil species despite rapid evolution. Importantly, most evidence for rapid adaptive evolution under natural conditions reported rather few evolving traits (e.g. Franks 2011; Nevo et al. 2012; Ravenscroftet al. 2015; Nguyen et al . 2016). Our findings caution that focusing on few evolving traits may overestimate the potential of rapid evolution for climate change adaptation.
High trait plasticity may further retard adaptive evolution (Shaw & Etterson 2012; Merilä & Hendry 2014; Kelly 2019), but this idea has rarely been tested in natural populations (Arnold et al. 2019). Our study, where the three rapidly evolving traits showed three contrasting magnitudes of plasticity (CV) indicates that plasticity and evolutionary potential are not necessarily related. However, this conclusion should be taken with caution because it is based on three partially correlated traits (Appendix, Fig. S3) which may have evolved in concert.
Our findings also provide little support for the idea that climate change leads to increased plasticity as a means to rapidly adjust the phenotype to novel conditions (Lande 2009; Arnold et al . 2019; Kelly 2019; Scheiner et al . 2020). Only a single trait, diaspore weight, had significantly increased plasticity under drought, and days to flowering showed a similar, non-significant tendency. Both responses, however, were opposite to the expected adaptive direction (e.g. later, not earlier flowering under drought; Appendix, Fig. S1), indicating non-adaptive plasticity (Acasuso-Rivero et al. 2019). Similarly, no clearly increased plasticity after drought was found by resurrection studies (Franks 2011; Hamann et al. 2018) and lower plasticity after CO2 elevation (Grossman & Rice 2014; but see Sultanet al. 2013 for increased plasticity during plant invasion). Thus, our study joins an –albeit small- body of equivocal evidence indicating that evolution of increased plasticity is no major pathway for climate change adaptation.
Overall, our study demonstrates that rapid evolution may play an important role for climate change adaptation in natural annual plant populations. The novel setup of our study – combining in situ climate manipulations with a natural climatic gradient and selection analyses under controlled conditions – provided independent, compelling lines of evidence that observed evolutionary shifts were adaptive. However, with rapid evolution in merely a subset of well-justified candidate traits, our study emphasizes the importance of multi-trait studies for assessing whether rapidin situ evolution may safeguard species under climate change.